CLEAR: Large-Scale Off-Road Planning with Terrain Abstractions

Pranay Meshram Charuvahan Adhivarahan Ehsan Tarkesh Esfahani Souma Chowdhury Chen Wang Karthik Dantu
Department of Computer Science and Engineering, University at Buffalo
Department of Mechanical and Aerospace Engineering, University at Buffalo Accepted for publication in IEEE Robotics and Automation Letters (RA-L), August 2026
University at Buffalo logo Drones Lab logo Office of Naval Research logo Moog Inc. logo

Video Presentation

CLEAR supplementary video showing the terrain abstraction and planning pipeline.

Paper Videos
CLEAR is a semantic-geometric terrain abstraction framework for large-scale off-road planning. It converts dense terrain maps into reusable, semantically aligned regions so planners can scale from local scenes to map-wide missions with reliable behavior.
CLEAR framework banner

Abstract

Long-horizon navigation in unstructured environments demands terrain abstractions that scale to tens of square kilometers while preserving semantic and geometric structure. Existing grids scale poorly, while quadtrees can misalign with terrain boundaries and omit landcover semantics that matter for traversability-aware planning.

CLEAR addresses the representation problem by constructing a reusable terrain abstraction from landcover and elevation. It produces convex, semantically aligned regions encoded as a terrain-aware graph. Across kilometer-scale maps, CLEAR reduces per-query planning time while preserving executability in physics-based simulation.

CLEAR: Terrain Abstraction

CLEAR operates through Boundary-Seed Decomposition, recursive plane fitting, and graph encoding. Boundary-Seed Decomposition places seeds in flat regions and along high-entropy semantic boundaries, then uses Voronoi partitioning to form compact convex regions. Plane fitting approximates elevation within each region and recursively subdivides regions whose error exceeds a tolerance. The final abstraction stores landcover, elevation geometry, convex boundaries, and adjacency for graph search.

CLEAR framework overview
CLEAR pipeline: landcover and elevation input, Boundary-Seed Decomposition, recursive planar surface fitting, and terrain-aware graph encoding.
CLEAR decomposition comparison
Compared with grid, hexagonal, and quadtree abstractions, CLEAR explicitly targets terrain boundaries and semantic-geometric consistency.

Results

CLEAR is evaluated on 9--100 km2 digital terrain maps with physics-based simulation. It reduces per-query planning time over unabstracted raw-grid A* after one-time abstraction and achieves 100% task completion across the evaluated maps in simulation.

CLEAR planning and simulation results
CLEAR planning and BeamNG simulation results on Wharton, Humphrey, and Rainier terrain maps.

BibTeX

@article{meshram2026clear,
  title={CLEAR: A Semantic-Geometric Terrain Abstraction for Large-Scale Unstructured Environments},
  author={Meshram, Pranay and Adhivarahan, Charuvahan and Esfahani, Ehsan Tarkesh and Chowdhury, Souma and Wang, Chen and Dantu, Karthik},
  journal={arXiv preprint arXiv:2601.13361},
  year={2026}
}